ACS Data Users Group

I noticed recently that the ACS 2019 - 2023 5-Year Summary File no longer provides block group-level estimates of population in group quarters via table B09019 (household type by relationship to head of household). However, these estimates remain available for previous vintages up to ACS 2022 5Y. Is anyone aware of changes in the availability of this table with the ACS 2023 5Y release?

I just replied to a similar question in this forum. In short, you can still get counts of group-quarters population for block groups by subtracting counts of population in households from total populations. Counts of population in households are available for block groups in B11002, B28005, B28008, and B28012, and maybe others.

Thanks, Jonathan! I was unaware of the additional tables that measured population in households, but that might do the trick. Apologies for duplicate posts.

So far so good for making the indirect estimates of group quarters population. However, I'm also trying to estimate standard errors based on Variance Replicate Tables for the two inputs (total population and population in households) and running into an issue. Since the two variables are dependent, I think we want:

SE(T – H) = SQRT(SE(T)**2 + SE(H)**2 – 2COV(T, H))

where T is total population and H is population in households.

However, this can result in negative values when SE(H) > SE(T), and I'm not sure how to handle those cases. These appear to be instances where the estimated group quarters population is greater than 0.

I haven't used the variance replicate tables but I use replicate weights for PUMS data all the time. I think that all you do is take the 80 x 2 tables (2 tables for each replicate) take the difference

difference i = total population table i - in households table i ( 80 differences)

then apply the formula to the 80 differences. See slide 38 in

https://www2.census.gov/about/training-workshops/2020/2020-02-12-acs-presentation.pdf

I'll dig into this in a minute and post something.

The variance replicate tables are generated using a jackknife type estimate:

en.wikipedia.org/.../Jackknife_resampling

Dave Dorer

PS to get the changes in 5 year tables between 2022 vintage and 2023 vintage look here

www.census.gov/.../5-year.html

Thanks, David! I just adapted this solution for the Variance Replicate Tables, and it seems to overcome the issues that I mentioned earlier.